Synthetic hedges, cointegration, mean reversion and similar stuff | Page 16 | Forex Factory

Options

Go to
Permalinks
Print Thread

Similar Threads

Simple Mean Reversion 621 replies

Pairs Trading: Reversion to the Mean 22 replies

Forward test of my new mean reversion strategy 11 replies

Question for synthetic and non-synthetic currency pairs 0 replies

Cointegration, Synthetic hedges, mean reversion in R, Tech Thread 0 replies

Attachments: Synthetic hedges, cointegration, mean reversion and similar stuff
Exit Attachments

Synthetic hedges, cointegration, mean reversion and similar stuff

  • #301
  • Mar 10, 2011 12:46pm Mar 10, 2011 12:46pm
Quoting Ccm168
Hi

Did you modified the 7 bit's code to run the ADF test? Would you mind sharing how you do it.

Thanks

Kim
There is Excel VBA code at link:
http://www.quantcode.com/modules/myd...%20desc&show=5

Input data is the difference in log prices of two pairs. Look for a very low value of p. 0.01 means unit root is rejected at 99% level. Any value of below 0.1 (90% of rejection) and higher order lag is good for "Government work" as OD pointed out.

Regards,

hnk
  • #302
  • Mar 10, 2011 2:44pm Mar 10, 2011 2:44pm
Quoting hnk
There is Excel VBA code at link:
http://www.quantcode.com/modules/myd...%20desc&show=5
Its possible (even recommended for arbomat) to do this with R, since in arbomat all data is already contained in variables in the running R session it would just need one or two more lines of code to do the ADF test (and one more line to print it nicely into the plot).

I'm just too lazy to implement it now (because I don't use it myself) but maybe somebody else here who has done it already could post the needed code, it should be trivial.
  • #303
  • Mar 10, 2011 11:12pm Mar 10, 2011 11:12pm
Quoting 7bit
Its possible (even recommended for arbomat) to do this with R, since in arbomat all data is already contained in variables in the running R session it would just need one or two more lines of code to do the ADF test (and one more line to print it nicely into the plot).

I'm just too lazy to implement it now (because I don't use it myself) but maybe somebody else here who has done it already could post the needed code, it should be trivial.
Hi 7bit,

First of all excellent work on your behalf for bringing this way of trading to FF. It will be nice to have ADF test built in your code as an option and will really help in identifying the mean-riveting pairs. I am a no programmer. All I can do is play with existing mq4 files a little bit and modify for different indicators or color. I myself have not done any work on R. If someone with familiarity with R can do it, it will definitely be an added bonus.

Regards,

hnk
  • #304
  • Mar 11, 2011 5:03am Mar 11, 2011 5:03am
I sent Steve Hopwood an EA that I have been developing with a partner that we can use with R.

The EA that I sent to Steve is very similar to your anti-grid/Snowball. The tools you have given us in this thread are invaluable. Thank you very much. I'm going to run the arbomat on as many pairs as I have time for during the weekends and see what I can come up with. I'd like to start a Skype group if anyone is interested.

I entered that Yen basket right before the big rally in yen (drop in Yen crosses) and it's doing quite well. lol. Thanks Old dog. (BTW, I use your pivots from your thread along with my own Fib grid and it has allowed me to take far fewer trades during the day and my average winner is 6x what it was before).
I hope you all take advantage of that NZDUSD USDCAD trade.

GL and GT
  • #305
  • Mar 11, 2011 12:53pm Mar 11, 2011 12:53pm
Quoting hnk
If someone with familiarity with R can do it, it will definitely be an added bonus.

Regards,

hnk
the R code is 2 lines:

ht <- adf.test(sprd, alternative="stationary", k=0)
cat("ADF p-value is", ht$p.value, "n")

just need to adapt/include them to the current mql files...

source: http://quanttrader.info/public/testForCoint.html
  • #306
  • Mar 18, 2011 11:16pm Mar 18, 2011 11:16pm
Hello 7Bit, thank you for your excellent work!

May be you should want to take a look at this video (specially at the middle of this)...

http://www.mathworks.cn/webex/record...507/index.html

It shows an example using cointegration to trade pairs, based on the Engle-Granger Method.

BR,
Fede
  • #307
  • Mar 19, 2011 1:17pm Mar 19, 2011 1:17pm
Hello,

during development of a backtesting engine in R, I realized an important thing that should be considered when evaluating spreads retrieved by whatever method (regression, PCA, ...).

Consider at time t0 we compute some coefficients, compute the corresponding spread, look at it and say "how nice a spread". This is extremely dangerous since this the computation of the coeffs incorporates historic information. We are not able to trade this spread along the time axis.

Instead, in order to do correct backtesting we need to re-compute the coeffs each trading step and look at the last value of the spread (including std, sharpe, etc.). If you do this over time the resulting series of "last-spread-values" do not look that nice.

In fact they do vary heavily at some times, so heavy that the spread becomes untradeble compared to the retrospective spread produced by the coeffs.

Just an observation I wanted to let you know.

Regards

p.s. The red line in the equity graphics is the begin of the japan crisis.
Attached Images
  • #308
  • Mar 19, 2011 2:49pm Mar 19, 2011 2:49pm
Showed the same before, the coefficients change over time.
There was also the example of changing the correlation in the motion in time.
  • #309
  • Mar 19, 2011 5:39pm Mar 19, 2011 5:39pm
Maybe you need forcast this coeff with GARCH models ?!

Regards
Ludo.

Quoting chrnbr
Hello,

during development of a backtesting engine in R, I realized an important thing that should be considered when evaluating spreads retrieved by whatever method (regression, PCA, ...).

Consider at time t0 we compute some coefficients, compute the corresponding spread, look at it and say "how nice a spread". This is extremely dangerous since this the computation of the coeffs incorporates historic information. We are not able to trade this spread along the time axis.

Instead, in order to do correct backtesting we need to re-compute the...
  • #310
  • Mar 23, 2011 11:04am Mar 23, 2011 11:04am
How does one change the grid spacing to 0.5 stddev for Arb-o-Mat ?
I looked in the expert already and was not able to figure it out.

Thank you...
  • #311
  • Mar 25, 2011 11:10am Mar 25, 2011 11:10am
// Rx("plot(curve, type='l', ylab=lbly, xlab=descr1, main='Arb-O-Mat', sub=descr2, col='cornflowerblue')");
// Rx("abline(lineb, linea, col='cornflowerblue', lty='dashed')");
// Rx("abline(lineb+stddev, linea, col='green', lty='dashed')");
// Rx("abline(lineb+2*stddev, linea, col='green', lty='dashed')");
// Rx("abline(lineb-stddev, linea, col='green', lty='dashed')");
// Rx("abline(lineb-2*stddev, linea, col='green', lty='dashed')");


Rx("plot(curve, type='l', ylab=lbly, xlab=descr1, main='Arb-O-Mat', sub=descr2, col='cornflowerblue')");
Rx("abline(lineb, linea, col='cornflowerblue', lty='dashed')");

Rx("abline(lineb+0.5*stddev, linea, col='green', lty='dashed')");
Rx("abline(lineb+1*stddev, linea, col='green', lty='dashed')");
Rx("abline(lineb+1.5*stddev, linea, col='green', lty='dashed')");
Rx("abline(lineb+2*stddev, linea, col='green', lty='dashed')");
Rx("abline(lineb+2.5*stddev, linea, col='green', lty='dashed')");


Rx("abline(lineb-0.5*stddev, linea, col='green', lty='dashed')");
Rx("abline(lineb-1*stddev, linea, col='green', lty='dashed')");
Rx("abline(lineb-1.5*stddev, linea, col='green', lty='dashed')");
Rx("abline(lineb-2*stddev, linea, col='green', lty='dashed')");
Rx("abline(lineb-2.5*stddev, linea, col='green', lty='dashed')");
  • #312
  • Mar 27, 2011 10:53pm Mar 27, 2011 10:53pm
Is there anyway we can show what the stddev is? The lines can be drawn, but I can't seem to get a numerical figure of what the stddev is. Thanks in advance.
  • #313
  • Mar 28, 2011 11:20pm Mar 28, 2011 11:20pm
Quoting 7bit
Its possible (even recommended for arbomat) to do this with R, since in arbomat all data is already contained in variables in the running R session it would just need one or two more lines of code to do the ADF test (and one more line to print it nicely into the plot).

I'm just too lazy to implement it now (because I don't use it myself) but maybe somebody else here who has done it already could post the needed code, it should be trivial.
Hi 7 bit

Why you said it is recommended to apply ADF Test for arbomat but not trend-o-mat, and what is the reason that you choose not to use ADF Test?

Thanks
  • #314
  • Apr 12, 2011 8:41am Apr 12, 2011 8:41am
Quoting Ccm168
what is the reason that you choose not to use ADF Test?
Laziness.

I believe these tests are a very valuable tool when automatically scanning though a huge number of possible baskets and finding the ones that are most promising but when I see one basket plotted in front of me with my own eyes then I think I can judge the quality of it much better than a single number from some test could do.
  • #315
  • Apr 12, 2011 8:45am Apr 12, 2011 8:45am
Quoting 7bit
This one looks interesting (the lines are 0.5 stddev, I just started running a grid strategy on it (buy on each level when it goes down, sell on each level when it goes up), wish me luck):
Attachment
This one has developed quite well so far :-)
Attached Image
  • #316
  • Apr 17, 2011 5:03pm Apr 17, 2011 5:03pm
It really is.
Seems like the most consistant upward move of the baskets history
http://i37.photobucket.com/albums/e9...DCADBasket.jpg
  • #317
  • Jun 2, 2011 9:57am Jun 2, 2011 9:57am
It has gotten pretty quiet arround here. How is everyone doing?
The NZDUSD/USDCAD basket is good for a sell right now.

I am also experimenting with non USD based currency paris, but I'm not sure if I'm doing it correct.
My approach is exactly the same, only that my inputs are some non usd pairs.
That way I found the following basket as an example(raw coefficients):
AUDCAD*-0.54
AUDCHF* 0.64
EURCAD* 0.36
EURCHF*-0.42
http://i37.photobucket.com/albums/e92/DKoH/basket-1.jpg

7bit, a while back you mentioned the following:
Quoting 7bit
After the regression the coefficients are multiplied by a certain factor to get lot sizes for dollar based lots. These factors are determined by the name of the instrument, it only works when there is "USD" in the name and it depends on whether it is xxx/USD or USD/xxx.

This is also the reason it does not work with crosses xxx/yyy without USD. It could be made to work but I simply did not because it is only an experimental proof of concept, a rough prototype, a quick and dirty hack, not more.
Does that mean, the whole approach of regression has to be different or is my basket from above valid and I only have to do the lot size calculation different?
If my approach is correct, how do I calculate the usd based lots from cross currencies?

Thanks!
Chris
  • #318
  • Jun 8, 2011 11:01am Jun 8, 2011 11:01am
Hi all!

I will to learn more on this topic, I will to download R for windows.
How is until now profitable this system/method?

Anybody is live ?

Thanks!


Quoting chrisx
It has gotten pretty quiet arround here. How is everyone doing?
The NZDUSD/USDCAD basket is good for a sell right now.

I am also experimenting with non USD based currency paris, but I'm not sure if I'm doing it correct.
My approach is exactly the same, only that my inputs are some non usd pairs.
That way I found the following basket as an example(raw coefficients):
AUDCAD*-0.54
AUDCHF* 0.64
EURCAD* 0.36
EURCHF*-0.42
http://i37.photobucket.com/albums/e92/DKoH/basket-1.jpg

7bit, a while back you mentioned the following:


Does that mean,...
  • #319
  • Jul 3, 2011 6:43am Jul 3, 2011 6:43am
Hi guys, here is another interesting R - quant trading blog
http://blog.quanttrader.org/
and a nice new book related to R based financial risk calculating
http://www.financialriskforecasting.com/

Another simple question: so far I havn't read in the entire thread an explanation for the "combined returns" axis in trend-o-mat. What does the sizing mean ? %, lots or something else ? When watching the Vimeo videos the combined return axis remains relatively constant (altering slightly) but the curve seems to grow steadily - any explanation for that ?
  • #320
  • Jul 5, 2011 11:51am Jul 5, 2011 11:51am
Quoting 7bit
Its possible (even recommended for arbomat) to do this with R, since in arbomat all data is already contained in variables in the running R session it would just need one or two more lines of code to do the ADF test (and one more line to print it nicely into the plot).

I'm just too lazy to implement it now (because I don't use it myself) but maybe somebody else here who has done it already could post the needed code, it should be trivial.
Why is ADF test ( http://www.stat.ucl.ac.be/ISdidactiq.../adf.test.html )recommended here ?

More appropriate should be the Johansen cointegration test for multiple variables. Check Wikipedia http://en.wikipedia.org/wiki/Johansen_test and the following R implementation http://rss.acs.unt.edu/Rdoc/library/...tml/ca.jo.html .
0 traders viewing now